One should look for what is and not what he thinks should be. (Albert Einstein)
Basics: Topic introduction
In this part of the course, we will cover the following concepts:
Introduction to neural networks
Neural networks use cases
Module completion checklist
Objective
Complete
Describe neural networks and identify their use cases
Discuss the architecture of a simple neural network and identify its components
Neural networks: introduction
Our brain uses neural networks to process information around us with the help of neurons that respond to the input from the environment
An artificial neural network (ANN), often called a neural network (NN), is a host of well-known machine learning algorithms that attempt to mimic the neural network a human brain uses
It uses interconnected nodes or neurons in a layered structure that resembles the human brain
Meet A.N.N
Let’s watch a video explaining how artificial neural networks work
Insight into preventing fatalities, based on data collected from car sensors
Use cases by algorithm type
Algorithm type
Use case
Classification
This image represents a horse; this email looks like spam; this transaction is fraudulent
Clustering
These two sounds are similar; these documents belong to the same topic / domain
Recommender systems
Given their web activity, this customer looks like they are going to stop using your service
Chat question: Can you think of use cases from projects that you have been working on?
Module completion checklist
Objective
Complete
Describe neural networks and identify their use cases
✔
Discuss the architecture of a simple neural network and identify its components
Simple neural network
Now that we’ve discussed the use cases of neural networks, we will walk through a simple neural network architecture and discuss its components
Note: Later in the course, we will go through a more complex neural network called deep learning
Perceptron
A perceptron, a fundamental concept in machine learning, is a type of artificial neural network
The image below depicts a perceptron, a neuron, connected with n other neurons
It is the simplest type of neural network and we will discuss each of the components in the subsequent slides
Neural networks: layers
The major components of a neural network are the:
inputs
weights
summation processor
activation function
outputs
Inputs
Weights
Summation processor
Before the summation processor “squeezes,” the computation goes through one or more hidden layers (this is where the black box concept of a neural network comes from)
A simple neural network will have up to two hidden layers, and anything over two hidden layers becomes a type of deep learning network
Activation function
There are many types of activation functions such as ReLU, sigmoid, tanh, and softmax
Based on the inputs, the function derives its current state, and outputs an activation value
Outputs
Neural networks: architecture
Note: There is no single NN architecture that works well for all problems!
The general architecture of a simple NN looks like a piece of layered cake
Depending on the number of hidden layers, layer type and its configuration (e.g., the number of neurons in it, the activation function and other parameters), each NN architecture might yield drastically different results for the same problem
Choosing the right NN architecture for the problem and tuning the model is instrumental to successful NN implementation
Deep learning vs. simple neural networks
Typically, one or two hidden layers are enough for most problems
Deep learning is a branch of machine learning that uses deep neural networks (DNNs)
DNNs are essentially the same as ANNs, with more complex architecture that calls for the use of multiple layers
Here is a diagram of how the two differ architecturally:
Knowledge check
Module completion checklist
Objective
Complete
Describe neural networks and identify their use cases
✔
Discuss the architecture of a simple neural network and identify its components